DIIN PAP SI - IZA Institute of Labor Economicsftp.iza.org/dp10773.pdf · IZA DP No. 10773...
Transcript of DIIN PAP SI - IZA Institute of Labor Economicsftp.iza.org/dp10773.pdf · IZA DP No. 10773...
Discussion PaPer series
IZA DP No. 10773
Terhi MaczulskijPetri Böckerman
Harsh Times: Do Stressors Lead toLabor Market Losses?
mAy 2017
Any opinions expressed in this paper are those of the author(s) and not those of IZA. Research published in this series may include views on policy, but IZA takes no institutional policy positions. The IZA research network is committed to the IZA Guiding Principles of Research Integrity.The IZA Institute of Labor Economics is an independent economic research institute that conducts research in labor economics and offers evidence-based policy advice on labor market issues. Supported by the Deutsche Post Foundation, IZA runs the world’s largest network of economists, whose research aims to provide answers to the global labor market challenges of our time. Our key objective is to build bridges between academic research, policymakers and society.IZA Discussion Papers often represent preliminary work and are circulated to encourage discussion. Citation of such a paper should account for its provisional character. A revised version may be available directly from the author.
Schaumburg-Lippe-Straße 5–953113 Bonn, Germany
Phone: +49-228-3894-0Email: [email protected] www.iza.org
IZA – Institute of Labor Economics
Discussion PaPer series
IZA DP No. 10773
Harsh Times: Do Stressors Lead toLabor Market Losses?
mAy 2017
Terhi MaczulskijLabour Institute for Economic Research
Petri BöckermanTurku School of Economics, Labour Institute for Economic Research and IZA
AbstrAct
mAy 2017IZA DP No. 10773
Harsh Times: Do Stressors Lead toLabor Market Losses?
This paper examines the effects of past stressful life events on subsequent labor market
success using data on twins matched to comprehensive register-based, individual-level
information on income and employment status. The long-term labor market outcomes
are measured during 20-year follow-up. We use the within-twin method to account for
unobservable family and genetic confounders. The twin design reveals three important
findings. First, stressors lead to worse labor market outcomes. Second, men are more
affected by financial and job-related stressors, while women are more affected by family
stressors. Third, the negative effects that stressors have on labor market outcomes diminish
as time passes.
JEL Classification: I31, J24, J31
Keywords: stressors, stressful life events, employment, earnings, co-twin control, twins
Corresponding author:Terhi MaczulskijLabour Institute for Economic ResearchPitkänsillanranta 3AHelsinki FI-00530Finland
E-mail: [email protected]
IZA DP No. 10773
non-Technical summary
mAy 2017
Life is full of stressors. Negative shocks include events such as job loss, divorce, widowhood
and the onset of major illness. Adverse life events may have long-lasting effects on an
individual’s ability to earn and be employed. We explore the relationship between past
stressful life events and long-term labor market success using a twin design. The earlier
literature in economic research has examined the effects of specific shocks, such as mass
lay-offs or the onset of divorce, on subsequent labor market outcomes. Our contribution
is that we used comprehensive measures for stressful life events that capture the full
spectrum of negative shocks that individuals are forced to cope with in their lives. Focusing
on single separate shocks does not account for this.
We use data on Finnish twins linked to comprehensive register-based, individual-level
information on earnings and employment status. The long-term labor market outcomes
are measured during 20-year follow-up. We exploit the within-twin dimension of the
linked data to fully account for both unobservable family and genetic confounders. This
is important, because literature has shown that family environment and genetics have
a profound role both in predisposing individuals to experience stressful life events, and
coping with these stressful events.
The twin design reveals three important findings. First, we find that those who have
previously experienced stressful life events have significantly weaker long-term labor market
attachment. Adverse shocks are also negatively linked to earnings for men and positively
linked to receipt of social income transfers for women. These findings are robust to using
comprehensive health-related controls. Second, there is a notable difference between men
and women regarding the importance of different types of shocks. Men are influenced by
work and financial events, whereas women are more likely than men to be distressed by
events within the family. And third, we find support for the adaptation hypothesis that
states that recent adverse life events matter more for subsequent labor market outcomes.
The results show that men adapt faster than women to negative life effects using labor
market success as a metric.
The fact that stressful life events profoundly matter for long-term labor market outcomes
provides support for the role of social insurance and other policies that accommodate
these shocks. The results are obtained within a Finnish setting. Importantly, the estimated
effects are economically significant despite that the Nordic welfare states provide extensive
programs to mitigate the negative effects of adversities on long-term labor market
outcomes. Therefore, our estimates likely constitute the lower bound for the size of the
effects that prevail in other countries.
2
I. Introduction
Shocks are a fact of life. Adverse life events, such as losing one’s job, profoundly affect
health behaviors such as alcohol consumption and smoking (McKee et al., 2003, Dawson et
al., 2005), subjective wellbeing at the individual level (Misheva, 2015) and labor market
outcomes (Jacobson et al., 1993). Shocks cause stress because they are often beyond an
individual’s control. Therefore, negative life events lead to potentially large welfare losses
that must be accompanied by the appropriate public policy responses.
Shocks come in many forms. Researchers have distinguished independent and
dependent life events (Bemmels et al., 2008). Some shocks, such as the death of a spouse or
other close relative are independent and outside of any one person’s control. However, other
shocks that individuals encounter in their lives, such as marital and financial problems, are
at least partially dependent on a person’s own behavior and autonomous choices.
The true effects of adverse shocks are challenging to identify for at least two reasons.
First, exposure to stressful life events may be influenced by shared environmental and
genetic confounders, and these factors may also be significantly correlated with labor market
success later in life. For example, early life conditions may exposure to shocks and have an
influence on earnings and employment outcomes. There is also evidence that exposure to
dependent life events is substantially influenced by genetic factors (Bemmels et al., 2008).
Second, individuals react in different ways to negative life events. Social support is
particularly important in the recovery process (Wethington and Kessler, 1986), and the
availability of such support varies significantly between individuals. Outside of social
support, there are many important psychological traits that help individuals to mitigate and
overcome the stress caused by adverse life events. These sources of resiliency include self-
confidence and autonomy, for example. Genetic factors play a significant role in explaining
3
human resilience to stress and adversity (Boardman, Blalock and Button, 2008; Waaktaar
and Torgensen, 2012).
This paper examines the effects of adverse life events on register-based long-run labor
market outcomes using twin design. There is an extensive stream in the economics literature
on the dynamic effects of various specific shocks on labor market outcomes, including
employment interruptions, such as mass lay-offs (e.g., Jacobson, LaLonde and Sullivan,
1993; Arulampalam, Gregg and Gregory, 2001; Korkeamäki and Kyyrä, 2014), the onset of
disability and health shocks (e.g., Currie and Madrian, 1999; Mok et al., 2008; García-
Gómez, 2011) and different types of household disruptions, such as divorce, widowhood or
sickness in the family (e.g., Haurin, 1989; Siegel, 2006). Most recently, Van den Berg,
Lundborg and Vikström (2017) examine the effect of an (exogenous) death of a child on
parents’ future labor market outcomes, marital status and health.
The intersection between psychology and other social sciences is increasingly fruitful
ground for new economic insights into policy-relevant issues. We use the Stressful Life
Events (SLE) index that systematically accounts for a broad set of adverse shocks. The
shocks described by the SLE index have been a locus of empirical research in the psychiatric
epidemiology literature. Adverse life events are stressors that have been shown to lead to
negative outcomes, such as the onset of a major depression in life (Kendler et al., 1999;
Tennant, 2002 for a literature review). Our novel contribution is that we introduce the SLE
concept to economic research and examine the effects of stressful life events on long-term
labor market attachment and earnings using a twin design.
A burgeoning literature shows that some primary shocks, such as lay-offs, may
predispose an individual to a series of secondary shocks, such as marital problems and risky
health behavior (Doiron and Mendolia, 2012; Black, Devereux and Salvanes, 2015), and that
individuals with poor health are more likely to become unemployed (Böckerman and
4
Ilmakunnas, 2009; Schmitz, 2011). The cumulative exposure to different shocks may have
substantial effects on labor market outcomes in the long run. Under this scenario, it is
difficult to disentangle the separate effects of a specific shock to subsequent labor market
losses when the total effects are partly influenced by other factors. A remedy to surmount
such worries is to use the SLE index because it captures the total burden of different types
of shocks in the long run. In a regression setting, it is in principle possible to simultaneously
control for a variety of different shocks and estimate the statistical significance of the
individual effects. However, the interpretation of the estimated effects may become
cumbersome if the regression is overloaded with many variables that have significant
interaction effects. Thus, the use of the SLE index mitigates residual confounding caused by
other shocks. The SLE index compactly summarizes information about several negative
aspects, which implies that we can combine different shocks into one index (or three
different indexes as we do in our paper) to create a single variable that provides an overall
account of the underlying structure of shocks.
We estimate the impact of the SLE index on long-term labor market outcomes using a
large and representative data on Finnish twins. Although the effects of various specific
shocks on labor market outcomes have been documented in the literature, the evidence on
these relationships using a twin design is relatively sparse. The previous literature has mainly
focused on the effects of birth weight on labor market outcomes (Black et al., 2007; Johnson
and Schoeni, 2011) and the impact of children on female labor supply and earnings (Silles,
2016). However, there is little evidence on the relationship between negative life events and
subsequent labor market outcomes using a twin design. The only exception is Lundborg,
Nilsson and Rooth (2014), who examine the effects of early life health on long-run
outcomes.
5
Using data on non-identical (dizygotic, DZ) twins allows us to account for two types
of shared family factors. First, there are adversities that two siblings within a family share,
such as the death of a parent or a grandparent. Second, it is important to address the role of
social support from the family or other shared social groups (such as a church) that help in
the recovery process. Using data on identical (monozygotic, MZ) twins allows us to further
control for inherited traits and preferences that are potential determinants of dependent
shocks that people face in their lives. Accordingly, by using MZ twins, we can control for
the individual differences in the resilience of adversities. The use of a single index for
negative shocks is particularly useful in a twin design because the sample sizes are relatively
small, especially for MZ twins.
To obtain a more nuanced picture, we distinguish independent and dependent life
events, as well as work and financial events and familial events. Twin data are linked to the
administrative information on long-term income and labor market attachment. Because we
analyze the effects in the context of a Nordic welfare state (Finland), we also examine the
effects of stressful life events on receiving social income transfers. To paint a dynamic
picture, we analyze the adaptation to stressful life events. This analysis is possible because
our data contain systematic information on the timing of various adverse shocks. Gaining
deeper knowledge regarding the adaptation to shocks is particularly useful for public policy
purposes. There is an apparent need for policy intervention if the effects of a shock on
subsequent labor market attachment and earnings are permanent.
The remainder of this paper is organized as follows. The next section describes the
Finnish twin cohort study that has been matched to register-based data on labor market
outcomes. This section also presents descriptive evidence on the heritability of adverse
shocks. The third section briefly discusses our empirical approach, and the fourth section
6
presents the baseline results of our analysis and various extensions. The final section
concludes the paper by putting our findings into the larger context of the literature.
II. Data
Twin survey and register data on labor market outcomes
Our analysis makes extensive use of the Finnish twin survey matched to the Finnish
Longitudinal Employer-Employee Data (FLEED). The linked data have been created for
research purposes by Statistics Finland. The data cover the period from 1975 to 2009. Our
twin survey sample is based on the Older Finnish Twin Cohort Study by the Department of
Public Health in the University of Helsinki, which was compiled from the Central Population
Registry of Finland (Kaprio et al., 1979; Kaprio and Koskenvuo, 2012). Initial candidates
for the survey were all Finnish twins born before 1958, identified using information on birth
date, the place of birth, sex, and surname at birth. The twin data contain only same-sex twin
pairs. A questionnaire was mailed to these candidates in 1975 to collect baseline data and to
determine their zygosity. Two follow-up surveys were conducted in 1981 and 1990.
The number of twin pairs in the data is 12,502, which corresponds to 25,004 individuals
(Kaprio et al., 1979). The twin study contains information on smoking, alcohol use,
symptoms of illnesses and reported diseases, medication use, physical characteristics,
psychosocial factors and multi-faceted information on experiences at work and in one’s
personal life. Based on previous explorations, our twin data are a representative sample of
the general population in Finland (Kaprio et al., 1979; Maczulskij, 2013a; Hyytinen et al.,
2013). The linked data remain representative also for the smaller sample of MZ twins
(Maczulskij, 2013b, p. 124-125).
7
The twin study is linked to the FLEED using personal identifiers (Hyytinen et al.,
2013). The FLEED consists of annual panel data over the 1990-2009 period. Using linked
data, we are able to comprehensively track the labor market behavior of those twins who
participated in the original twin surveys. FLEED is based on various administrative registers
on individuals and firms that are collected and/or maintained by Statistics Finland. The data
include information on an individual’s exact labor market status and income taken directly
from tax and other administrative registers. Thus, the income and employment information
do not suffer from the characteristic shortcomings of survey data (e.g., underreporting, recall
errors or top-coding).
We focus on the non-retired primary working-age persons, who were at least 33 years
old in 1990. The twin survey in 1990 was mailed only to twin pairs born 1930–1957 (n =
12,450 individuals) with the response rate of 77%. Our analysis focuses on twin pairs for
whom we have data on experiencing stressful life events, relevant covariates, and labor
market outcomes. After excluding missing information further decreases the sample size to
6,247 twin pairs. Those observations that do not have information on one’s sibling are also
excluded from the estimation sample, resulting 3,216 twin pairs (i.e., 6,432 individuals). Of
these individuals, ~58% are females and ~37% are MZ twins.
Outcome measures
As the outcome variables from FLEED, we use employment, earnings and social income
transfers. Our measure for employment months is calculated as the average number of
employment months per year over the 1990-2009 sample period. We also use two income
measures. First, we approximate the lifetime earnings by the logarithm of the average of
annual wage and salary earnings and self-employment income over the 1990-2009 period.
8
Second, we use social income transfers. Specifically, the data contain information on total
annual taxable income obtained from the Finnish tax authorities. Total income is a broader
concept than earnings because total income also includes income transfers and social
security benefits, such as parental leave and unemployment benefits. Annual social income
transfers are calculated by subtracting annual wage and salary earnings and self-employment
income from total annual taxable income. Lifetime income transfers are measured by the
logarithm of the average social security benefits and income transfers over the 1990-2009
period. Both income measures and social income transfers are deflated to 2009 euros using
the consumer price index of Statistics Finland.
The sufficient condition for a twin pair to be included in the analyses is that we observe
the labor market outcomes for the twin pair at least once (one year) during the observation
window of 1990-2009. Thus, we do not make the assumption that everyone in the sample
should be working and have positive earnings for the entire 20-year time period.
Accordingly, when an individual becomes retired, his/her subsequent person-year
observations are excluded from the calculation of labor market outcomes.
Assessment of stressful life events
The SLE index is measured by the weighted sum of experiencing negative life events using
self-reported data from the 1990 survey. The twin data contain a 17-item Holmes and Rahe
life event inventory. The twins were requested to indicate which SLEs they had experienced
and to specify the timing of the events as follows:
1 – ‘Never’
2 – ‘During the last six months’
9
3 – ‘During the last five years (excluding the events during the first 6 months)’
4 – ‘Happened to me earlier’
A thorough description of the construction of the SLE index is provided in earlier
publications (Lillberg et al., 2003; Riese et al., 2013) that used the twin data from the 1981
survey. Of the 17 items, 11 were initially rated as negative (Riese et al., 2013). There was
some disagreement on two items as being negative (‘Marked increase in work load’ and
‘Marked change in the health of a family member’). However, Riese et al. (2013) found that
the results were not sensitive to the manner in which they calculated the SLE index, whether
they used 9 or 11 items of negative SLEs. This finding is important because we use data
from the 1990 survey for SLEs that do not include the question that reveals whether an
individual has experienced a ‘Marked increase in work load’. According to Riese et al.
(2013), using 10 items instead of 11 to calculate the SLE index should not have an impact
on our main results.
The SLE index was calculated as the weighted sum of these 10 items. Prior findings
suggest that the impact of life events at a low frequency is larger compared with those at a
high frequency (Masuda and Holmes, 1978). The weights for the SLEs were calculated as
the inverse of the lifetime prevalence (1 minus prevalence) of each negative SLE within our
sample.1 The prevalence was defined as ever having experienced the specific SLE. Those
subjects who had more than two items missing were excluded. Analogously to Riese et al.
(2013), if subjects had less than three missing SLE items, then they were coded as ‘never
1 This method closely tracks the use of ‘Life Change Unit’ (LCU) weights that originate from the Social
Readjustment Rating Scale (SRRS) (Holmes and Rahe, 1967). These weights are measured on the basis of the
extent to which particular SLEs are assumed to require adaptive behavior. For example, widowhood has the
highest weight (100). We use this weighting method in the robustness tests.
10
experiencing’. The SLE index was standardized to have a mean of zero and a standard
deviation of one to obtain easily comparable regression coefficients. The prevalence of each
SLE and their weights are documented in Table 1. The non-normalized frequency histogram
of the SLE index is presented in Figure 1.
[Table 1 and Figure 1 in here]
Two relevant empirical facts have been established in the literature. First, previous studies
have shown that environmental and genetic factors contribute differently to the variance of
dependent and independent life events. In particular, dependent life events are explained by
genetic factors for the most part, whereas shared and unshared environmental factors are a
larger contributor to the variance in independent life events (Bemmels et al., 2008). Second,
men and women are distressed by distinctly different types of adverse events. Men are
primarily more influenced by work and financial events, whereas women are more likely
than men to be distressed by various (social) network events and shocks within the family
(Kessler and McLeod, 1984; Conger et al., 1993; Kendler et al., 2001). This finding is
consistent with the “cost-of-caring” hypothesis according to which the greater vulnerability
of women is explained by a higher emotional engagement in others’ lives. Using these
stylized empirical facts from previous studies, stressful life events were further categorized
into three non-overlapping classes.
Specifically, events were measured by the weighted sum of exposure to negative life
events as follows:
1 – Dependent work and financial events: Loss of a job, difficulties with a boss or
colleagues at work and financial difficulties
11
2 – Dependent familial events: Divorce or separation, difficulties with a spouse, sexual
difficulties, and disease or injury leading to more than three weeks of disability from
work
3 – Independent familial events: Death of a spouse, death of a close relative or friend, and
change in the health of a family member
There may be disagreement about ‘Disease or injury causing over three weeks of disability
from work’ being included in the class of dependent familial events. For example, being
involved in an accident may be outside the control of an individual person. However, our
results are robust to the use of differently categorized classes of events, in which we removed
the ‘Disease or injury causing…’ event from the category of dependent familial events and
included it to the class of independent familial events.
Controls
We control for socio-economic characteristics, the number of diseases and previous wage
level in all specifications. The socioeconomic confounders include age and age squared,
education (measured in years, based on the highest completed education level) and marital
status (1 if ever married, as reported in the 1975, 1981 and 1990 twin surveys). The number
of chronic diseases (1981) is used to account for the pre-existing health endowment. The
chronic diseases include, among others, emphysema, chronic obstructive pulmonary disease,
high blood pressure, angina pectoris, peptic ulcer, diabetes, and gout.
We account for the possibility that the relationship between stressful life events and
subsequent labor market outcomes is driven by reverse causality. Early income is strongly
12
correlated with subsequent labor market success. Thus, if early labor market success or
failure has affected the experience of negative shocks, then our estimates might reflect
reverse or two-way causality, at least in part. Our measure for early labor market success is
the individual’s annual taxable income in 1980.2 This information originates from the
comprehensive Longitudinal Population Census by Statistics Finland to which the twin data
have been linked with personal IDs.
There are several potential mechanisms between experiencing stressful life events and
subsequent labor market success. We explore these mechanisms by adding measures for the
crucial aspects of health behavior, as well as the measures for mental stability in the
additional models as covariates. For example, adverse life events have been found to affect
risky health behavior, such as excessive alcohol consumption and smoking (McKee et al.,
2003, Dawson et al., 2005), which lead to substantial losses in the labor market (Böckerman
et al., 2015a; Böckerman et al., 2015b). The health-related controls include smoking and
alcohol use. Smoking is measured using pack-years in 1990, which describes lifetime
consumption of cigarettes (Böckerman et al., 2015b). Our measure of alcohol use is the
extreme case of binge drinking, which is based on the question regarding the pass out
frequency during the past 12 months in the 1990 twin survey.
Mental stability is measured using the indicators of neuroticism and extraversion that
originate from the 1981 survey. We add neuroticism as an additional control because there
is a previously established link between experiencing adverse shocks and neuroticism on the
one hand (Kendler et al., 2012; Riese et al., 2013) and between neuroticism and labor market
2 Some of the events may have happened before the initial health endowment and wage level are measured.
However, the SLE index is measured in 1990, and most of the events have happened during the last five years,
according to the data. The events must have happened over ten years ago for the controls not being pre-
determined.
13
success on the other hand (Nandi and Nicoletti, 2014). In turn, extraversion may predispose
individuals to experience negative life events more positively (Lucas et al., 2000).
Neuroticism (extraversion) was assessed by 10 (9) items in the short form of the Eysenck
Personality Inventory. We also add the use of tranquilizers from the 1990 survey as a
covariate, which captures an aspect of mental health. Tranquilizer use has the value of one
if the twin reports using a positive quantity of tranquilizers in 1990.
Heritability of stressful life events
Table 2 reports the intra-class correlations of the SLE index and its three classes between
DZ and MZ twins. The within-pair correlation of the SLE index is 0.13 for DZ twins and
0.24 for MZ twins. Therefore, MZ twins are much more similar with respect to each other
than DZ twins are in their reporting of adverse life events. This pattern is more striking when
the dependent and independent life events are analyzed separately. We find that there is no
significant discrepancy between the intra-class correlations of independent familial events
between DZ and MZ twins (0.08 vs. 0.09). This observation most likely reflects the fact that
random shocks are beyond one’s own control.
[Table 2 in here]
The results suggest that exposure to negative life events is partly explained by genetic
factors. The contribution of heritability is more profound for dependent adverse events. We
evaluate this pattern further using the DF-model (DeFries and Fulker, 1985), which yields
estimates for the shared environment and heritability of SLE using the following equation
estimated by using OLS (Ordinary Least Squares):
14
𝑆𝐿𝐸1𝑗 = 𝛼0 + 𝛽1𝑆𝐿𝐸2𝑗 + 𝛽2𝑅𝑗 + 𝛽3𝑅𝑗𝑆𝐿𝐸2𝑗 + 𝜀1𝑗 , (1)
where 𝑆𝐿𝐸1𝑗 is the SLE index for twin 1 in family j, 𝑆𝐿𝐸2𝑗 is the SLE index for twin 2 in
family j, and R is the genetic relatedness (0.5 for DZ twins and 1 for MZ twins). Thus, the
variation in experiencing stressful life events is decomposed into components that are
attributed to a shared environment (coefficient 𝛽1) and genetic effects (coefficient 𝛽3). The
intra-correlation of the outcome variable within MZ twins is, in certain cases, more than
twice of that for the DZ twins, i.e., 𝑟𝑀𝑍 > 2𝑟𝐷𝑍. This pattern suggests that additive genetic
effects may be present, and the model can yield estimates that fall within the categories 𝛽3 >
1 and/or 𝛽1 < 0 (Waller 1994). In this setting, the basic DF-model can be re-formulated as:
𝑆𝐿𝐸1𝑗 = 𝛼0 + 𝛽2𝑅𝑗 + 𝛽3𝑅𝑗𝑆𝐿𝐸2𝑗 + 𝛽4𝐷𝑗𝑆𝐿𝐸2𝑗 + 𝜀1𝑗, (2)
where D is 0.25 for DZ twins and 1 for MZ twins. Broad-sense heritability is the sum of the
parameter estimates 𝛽3 + 𝛽4 and corresponds to the heritability estimate (genetic effect) 𝛽3
in equation (1). Thus, this specification omits the term of shared environment, i.e., we set
𝛽1 = 0. In both specifications, the double-entry method is used (Cherny et al., 1992), which
means that each twin is entered twice in the model: once as proband and once as co-twin. In
accordance with Kohler and Rodgers (2001), we calculate the asymptotic standard errors for
double-entry twin data.
The estimates for the shared environment and genetic heritability are reported in Table
2 (Columns 3 and 4). The estimated contribution of heritability in the SLE index is 0.23. The
estimate for the shared environment is notably lower at 0.005 and statistically insignificant.
15
Under the standard assumptions of the DF-model,3 the results show that the variation in
exposure to adverse shocks is heritable at a rate of 23%. In the case of dependent work and
financial events, the estimate for the shared environment is negative, which indicates the
presence of additive genetic effect. The intra-class correlations reveal the same pattern. The
correlation within MZ twins (0.34) is more than twice the correlation for DZ twins (0.08).
The result of our preferred model (equation 2) shows that the heritability is 26%.
Interestingly, neither the genetic effects nor the shared environment appear to explain the
variation in dependent familial events. The variation in independent familial events (such as
the death of a close relative) is statistically significantly explained by the shared environment
and the effects of heritability are statistically zero. These results support the external validity
for our estimates because Bemmels et al. (2008) found similarly that the variance in
dependent life events is significantly explained by genetic factors, whereas the shared
environmental effects are the largest contributor to the variance in independent familial
events.
3 The DF-model is based on four key assumptions: 1) genes and the environment have additive effects, 2)
additive environmental influence is similar for DZ and MZ twins, 3) there is no assortative mating, and 4) there
is no correlation or interaction between shared environment and genetic factors (e.g. Behrman and Taubman,
1976). A discussion of the DF-model and criticisms of it are presented in Maczulskij (2013a).
16
III. Statistical method
Our main econometric analysis is based on the following model:
𝑌𝑖𝑗 = 𝛼 + 𝛽′𝑆𝐿𝐸𝑖𝑗 + 𝑓𝑗 + 𝑔𝑖𝑗 + 𝜀𝑖𝑗 (3)
where 𝑌𝑖𝑗 is the long-term labor market outcome of twin i in twin-pair j. 𝑆𝐿𝐸𝑖𝑗 is the index
for past adverse shocks in life, 𝑓𝑗 is the unobserved family endowment common to both twins
of pair j, 𝑔𝑖𝑗 is the unobserved genetic endowment specific to twin i of pair j, and 𝜀𝑖𝑗 is a
random shock to twin i of pair j.
The equation is first estimated by OLS using cross-sectional variation between
individuals. This model provides an estimate for 𝑆𝐿𝐸 that is denoted by 𝛽𝑂𝐿𝑆. Because the
SLE index is standardized, 𝛽𝑂𝐿𝑆 measures in percent terms how much one standard deviation
increase in the SLE index is associated with an increase/decrease in a specific long-term
labor market outcome 𝑌𝑖𝑗. For 𝛽𝑂𝐿𝑆 to be a consistent estimator of the coefficient of 𝛽, the
moment condition 𝐸[𝑓𝑗 + 𝑔𝑖𝑗 + 𝜀𝑖𝑗|𝑆𝐿𝐸𝑖𝑗] = 0 should hold. This condition does not hold if
𝑓𝑗 or 𝑔𝑖𝑗 are correlated with the SLE index. Because 𝑓𝑗 and 𝑔𝑖𝑗 are generally not accounted
for in observational data, the omission of these terms yields biased estimates for the
association between SLE and subsequent labor market success. For example, a positive
correlation between risk-loving behavior and dependent SLEs, such as having a divorce or
being involved in an accident that causes injuries, will lead 𝛽𝑂𝐿𝑆 to overestimate the true
value of 𝛽.
We use within-twins variation among the DZ twins to difference out the family effects,
𝑓𝑗. In the twin-differenced DZ sample, the estimator is consistent if 𝐸[(𝑔2𝑗 − 𝑔1𝑗) +
(𝜀2𝑗 − 𝜀1𝑗)|(𝑆𝐿𝐸2𝑗 − 𝑆𝐿𝐸1𝑗)] = 0, where the terms inside the brackets refer to the within-
17
sibling differences of the variables. The condition does not hold if (𝑔2𝑗 − 𝑔1𝑗) is correlated
with (𝑆𝐿𝐸2𝑗 − 𝑆𝐿𝐸1𝑗). Furthermore, if the twins are identical, (𝑔2𝑗 − 𝑔1𝑗) = 0. Thus, the
genetic effects can also be differenced out. Thus, using within-twins variation among the
MZ twins yields an estimator that is denoted by 𝛽𝑀𝑍. If shocks in life are random conditional
on genetic endowment, then 𝛽𝑀𝑍 is a consistent estimate of 𝛽.
There are four possible problems with the twin-based design. First, there is a potential
endogeneity problem caused by omitted variables if there are unaccounted variables that
affect both adverse life events and subsequent labor market outcomes. Because independent
life events are considered to be truly exogenous, this omitted variable bias should be relevant
solely in the case of dependent life events. For example, MZ twins can differ in their initial
endowments, such as birth weight (Bound and Solon, 1999). Low birth weight has been
linked to various adult outcomes, such as lower cognitive ability, lower mental stability (i.e.,
neuroticism), deficits in social skills (introversion), lower autonomy, lower probability of
mating, and poorer labor market outcomes (e.g., Behrman and Rosenzweig, 2004; Black,
Devereux and Salvanes, 2007; Kajantie et al., 2008; Eryigit-Madzwamude et al., 2015). If
low birth weight is positively related to experiencing dependent (adverse) life events, then
our within MZ twin-pair results would be upward biased because we have no information
on birth weight. However, lower mental stability (such as neuroticism) may capture, at least
partly, the potential negative effects of low birth weight on both experiencing dependent
(adverse) life events and labor market success.
The second problem is that twin-differencing may exacerbate the measurement error
problem compared to a conventional cross-sectional analysis (Griliches, 1979; Bound and
Solon, 1999). If life event measures were subject to classical measurement error, then our
results would be downward biased and lead to conservative estimates for adverse life events.
The third potential problem is that SLEs can also happen during the 1990-2009 window.
18
These post 1990 SLEs that are omitted in the linked data thereby potentially confound
estimates. It is not unreasonable to think that someone who is 33 in 1990 would have such
an event (divorce, death of a parent, spouse or child) over the next 20 years and that that
event would affect his or her labor market experience. However, it is typical to exclude the
additional shocks later in life also in the earlier literature that has examined the effects of
specific shocks on labor market outcomes, and some of these later shocks may also be
endogenous with respect to the labor market status. The fourth potential problem is that the
SLE index accounts only for negative shocks by construction. It is possible that there are
also positive shocks that counter negative ones, buffering the effects on labor market
outcomes over the 1990-2009 period. This would imply that we obtain conservative
estimates for the effects of negative shocks.
IV. Results
Descriptive evidence
Table 3 documents the mean values of the variables by gender. We also report F-test statistics
for the null hypothesis of equal group means in column 3. The means of the variables are
consistent with well-known empirical facts. Women have higher scores in the SLE index.
Interestingly, this pattern is driven by women experiencing more adversities within the
family. Women have weaker labor market success in terms of long-term earnings and
employment compared with men; however, they receive less social income transfers over
time. Although women drink and smoke less, they have more chronic diseases, and they also
use more tranquilizers. Women have higher scores in neuroticism (e.g., Flecher, 2013),
whereas men have higher scores in extraversion.
19
The means of the absolute values of the twin differences in the MZ sample are reported
in columns 4 and 5. These statistics show that there is a sufficient amount of within-twin
pair variation in the data even among MZ twins, which is a necessary condition for model
identification. Therefore, our results do not rely on an idiosyncratic subset of the sample of
twins with unusual differences.
type of stressful life event index and individual’s basic individual characteristics by
gender. The individual characteristics are measured in 1980/1981. Therefore, they are
arguably pre-determined for our stress measures to a large extent. The correlations are
reported in Table A1 in the Appendix. The results show that the within-MZ differences in
initial labor market status (unemployment) and skill-level (wages in 1980 and education
level) do not explain the differences in experiencing stressful life events for men. For women
we find that previous wage level is positively related to experiencing independent familial
events. Personality characteristics are also important in explaining differences in
experiencing dependent stressful life events for both genders. The number of chronic
diseases and excess alcohol use in 1981 are positively related to experiencing dependent
work and financial events for men. For women we find a strong relationship between risky
health behaviors and all types of negative events, also regarding independent familial events,
such as death of a spouse and illness in the family. This pattern may be explained by non-
random mating and convergence (Ask et al., 2012).
[Table 3 in here]
20
Main results
The effects of stressful life events on long-term earnings, employment months and social
income transfers are reported in Table 4 for men and in Table 5 for women. The
specifications marked with ‘A’ report the estimates for the SLE index, whereas the
specifications marked with ‘B’ report the estimates for three non-overlapping classes of
SLEs: dependent work and financial, dependent familial, and independent familial events.
The controls include marital status, education years, the initial number of chronic diseases
(1981) and the previous earnings level (1980). The OLS specification (column 1) also
controls for age to be comparable with the specifications (columns 2-4) that are estimated
using the within-twin pair regressions that automatically account for such an invariant
within-twin variable.
We first discuss the results for men. The baseline estimates that use the standard OLS
specification reveal that stressful life events are negatively correlated with both long-term
earnings and labor market attachment. Negative life shocks are also positively linked to
receiving social income transfers over the estimation window. The estimates are
economically significant. The estimates show that a one-standard deviation increase in the
SLE index is associated with a reduction in average employment months of ~0.5. This
decrease corresponds to 10 months over our 20-year observation period. A similar increase
in the SLE index is associated with a decrease in average earnings of 9% and an increase in
social income transfers of 42%.
The point estimates tend to decrease when we focus on the twin-differenced DZ-MZ
model (column 2) and the DZ model (column 3), which both control for shared environment.
The overall pattern of the estimation results nevertheless remains the same. The results for
the MZ sample (column 4) confirm our earlier findings for earnings and employment when
21
both shared environmental and genetic factors are controlled for. These preferred estimates
reveal that a one-standard deviation increase in the SLE index is associated with a decrease
in average employment months of ~0.3 and average earnings of ~5%. To further illustrate
the quantitative magnitude of the estimated effects, one-standard deviation increase in the
SLE index is roughly equivalent to two additional events of average prevalence or one
additional event of low prevalence.
Table 5 reports the estimates for women. The baseline OLS estimates (column 1) are
comparable with those for men in Table 4. The results remain unchanged when we focus on
the twin-differenced models (columns 2 and 3) that account for shared environmental
factors. Using earnings as the outcome variable, our preferred twin-differenced MZ model
(column 4) estimate shows that experiencing stressful life events is no longer associated with
lower earnings for women when both shared environment and genetic factors are fully
controlled for. The estimate for social income transfers in the MZ sample, however, remains
statistically significant at 0.34. This point estimate implies that a one-standard deviation
increase in the SLE index is associated with an increase of receiving social income transfers
by 40%. A similar increase in the SLE index is associated with a decrease in average
employment months by ~0.3.
The separate estimates for the three classes of the SLE index are reported in the
specifications marked with ‘B’ in Tables 4-5. Our preferred results for the MZ sample that
use earnings as the outcome variable show that men are adversely influenced by work and
financial events, whereas women are not distressed by work-related shocks. Using
employment as the outcome variable the quantitative magnitude of work-related shocks is
also significantly larger for men compared with that for women. Interestingly, experiencing
adverse independent familial events, such as the death of a spouse or sickness in the family,
is associated with receiving less social income transfers for men and more for women. A
22
possible explanation for this observation is that work-oriented men may seek support from
social networks from their workplaces, which induces them to work more and implies a
lesser need for social income transfers. It is also possible that exogenous family shocks lead
to a notable increase in social income transfers for women to compensate for lost income
because men are usually the primary family breadwinners (cf. Bargain et al., 2012). Thus,
men and women respond differently to negative shocks. Our findings are in accordance with
the results in Kendler et al. (2001) who showed that men are more sensitive to work-related
problems, whereas women are more sensitive to various network events, including the death
of a spouse.
When we use income transfers as our outcome variable, the estimate of SLE index is
highly positive in the DZ sample but not in the MZ sample for men. This suggests that some
of the genetic factors are positively correlated with experiencing adverse shock, and
especially dependent familial shocks, which could lead to an upward bias in the OLS and
within-DZ estimates. One explanation is risk preferences, because risk-loving behavior is
positively related to having a divorce (and other problems with a spouse) and disability
(e.g.,Light and Ahn, 2010). However, the reverse is true for women: the estimate for
experiencing adverse (independent familial) events is higher in the MZ sample than in the
DZ sample when we use income transfers as the outcome variable. Because independent
familial events are exogenous, and thus independent on genetics, the difference in the DZ
and MZ estimates could be explained by differences in the resilience of shocks.
We also took into account that individuals in our sample can only experience the death
of a spouse, divorce and marital discord if they are married. To this end, we restricted our
sample to married individuals only. There may also be a correlation between age and the
type of shock, e.g., losing a spouse after the age of 45 when there are meaningful differences
in the marriage market. Both of these additional tests provided highly comparable results for
23
women. For men, we observed that the estimates were, for the most part, quantitatively
similar, but not always statistically significant due to smaller sample sizes (results not
reported).
[Tables 4-5 in here]
Robustness checks
To explore the sensitivity of the main results, we have estimated additional specifications.
We briefly discuss each of these results.
The main estimation results are based on the standard formulation of the SLE index.
However, we have considered the robustness of our estimation results regarding the exact
weighting method for the SLE index (results are not reported). First, we set the threshold to
zero for missing items. The number of observations decreased from 6,432 to 5,460. The
estimates were quantitatively similar, but sometimes only marginally significant.
Importantly, the statistical insignificances were not driven by the smaller (absolute) point
estimates but a smaller sample size. Second, the results were robust to the use of equal
weighting method. Third, we used the weights for an individual SLE from the Holmes and
Rahe (1976) scale, i.e., the LCU weights (see the weights in Riese et al., 2013, Table 1).
These results were highly comparable with the main results, except that the estimates also
showed larger effects of familial events on long-run earnings for women.
We have excluded the key control variables from the models (marital status, education,
initial health endowment and the previous wage level). The results are stable and the earlier
conclusions remain intact. The only exception is that when the previous wage level is
excluded from the set of controls, there is a stronger relationship between familial events
and labor market outcomes for women. For example, dependent familial shocks (such as
24
divorce) contribute positively on females’ earnings. This result is likely to be driven by
reverse causality between females’ earnings level and subsequent familial events, such as
marital difficulties. This is in line with the results by Johnson and Skinner (1986), who show
that women increase their labor supply several years prior to separation.
Additional aspects
We have also examined the role of risky health behavior (alcohol consumption and smoking)
and mental stability (neuroticism, extroversion and the use of tranquilizers) as potential
mediators in the relationship between stressful life events and subsequent labor market
outcomes. This examination is an important extension of the earlier literature because stress
may trigger changes in substantial health behavior, such as excessive alcohol consumption
(McKee et al., 2003; Dawson et al., 2005), which leads to serious difficulties in the labor
market (Böckerman et al., 2015a). We use the within-twin pair variation for these variables
to explore the robustness of our within twin results for three classes of the SLE index. The
earlier results for the SLE indexes remain intact, showing that the negative effects of adverse
shocks on labor market outcomes are not primarily caused by health behaviors.4
Finally, we used an alternative measure for (weak) labor market attachment, namely
the average number of unemployment months. Our preferred within-MZ results show that
stressors are positively related to unemployment months in the long run and that this
relationship is entirely explained by experiencing adverse work and financial events. The
4 We do not include the measures for risky health behavior as controls in the baseline models because smoking
and drinking are measured in 1990 based on recall (with likely measurement error) and are also likely to change
over the 20-year time span used for labor market outcomes. Moreover, changes in health behavior between
1990-2009 could be endogenously related to unobserved SLEs during this time period.
25
estimate is approximately 0.29 for both genders, indicating that a one-standard deviation
increase in the stressors related to work and financial events is associated with an increase
in average unemployment months by ~0.3 (not reported in tables).
Timing of stressful life events and labor market outcomes
Hedonic adaptation refers to the psychological process in which individuals return to their
earlier baseline level of happiness following a change in external life circumstances.
Misheva (2015) found that more recent traumatic events, such as an assault or rape, have a
much greater impact on various aspects of emotional well-being. Another interesting study
is by Clark et al. (2008) who provided evidence on the adaptation hypothesis for
experiencing several life events, such as divorce, widowhood and layoff. Using German
panel data that allows Clark et al. (2008) to follow individuals over time, they have also
reported that there is an incomplete adaption to unemployment for men.
We analyze the adaptation to stressful life events using labor market success as the
outcome variable. This analysis is possible because our twin data contain systematic
information on the timing of various adverse shocks. We used this information to distinguish
between recent (the last six months), later (during the last 5 years excluding the latest six
months) and distant (occurred over five years ago) SLEs as measured by the weighted sum
of the events.
The estimates from our preferred within MZ twin-pair specification are reported in
Table 6. These results confirm the overall validity of the adaptation hypothesis. It appears
that more recent adverse events matter more for men. Women are also affected by the events
that occurred to them during the last five years. For both men and women, none of the distant
events matter for subsequent labor market success.
26
We have estimated models that examine the adaptation to shocks also using the three
non-overlapping indexes because individuals may adapt to different shocks in different
amounts of time. The results using the long-term labor market outcomes with the separate
classes of adverse shocks are broadly consistent with the patterns using the overall SLE
index (not reported). The only exception is that more distant independent familial events are
associated with receiving less social income transfers for men.
[Table 6 in here]
5. Conclusions
Life is full of stressors. Negative shocks include events such as job loss, divorce and the
onset of major illness. Adverse life events may have long-lasting effects on an individual’s
ability to earn and be employed. This paper explores the relationship between past stressful
life events and long-term labor market success using a twin design. The earlier literature in
economic research has examined the effects of specific shocks, such as mass lay-offs or the
onset of divorce, on subsequent labor market outcomes. Our contribution is that we used
comprehensive measures for stressful life events that capture the full spectrum of negative
shocks that individuals are forced to cope with in their lives. Focusing on single separate
shocks does not account for this.
We use data on Finnish twins linked to comprehensive register-based, individual-level
information on earnings and employment status. The long-term labor market outcomes are
measured in adulthood. To identify the effects, we use twin data because the literature has
shown that family environment and genetics have a profound role in predisposing
individuals to experience stressful life events in certain ways. Thus, we exploit the within-
27
twin dimension of the linked data to fully account for both unobservable family and genetic
confounders.
Our main finding is that stressful life events are an important, but neglected,
determinant of long-term labor market outcomes. Using within-twin pair estimations for
monozygotic twins, we find that those who have previously experienced stressful life events
have significantly weaker long-term labor market attachment. Adverse shocks are also
negatively linked to earnings for men and positively linked to receipt of social income
transfers for women. These findings are robust to using comprehensive health-related
controls.
We also establish two other important empirical patterns. First, there is a notable
difference between men and women regarding the importance of different types of shocks.
Men are influenced by work and financial events, whereas women are more likely than men
to be distressed by events within the family. Second, people adapt to shocks. We find support
for the adaptation hypothesis that states that recent adverse life events matter more for
subsequent labor market outcomes. The results show that men adapt faster than women to
negative life effects using labor market success as a metric.
The fact that stressful life events profoundly matter for long-term labor market
outcomes provides support for the role of social insurance and other policies that
accommodate these shocks. The results are obtained within a Finnish setting. Finland is a
much smaller, more culturally homogenous country with a more robust welfare state than
some other EU countries or the US. We clearly need more evidence on the impact of stressful
life events in other cultural and institutional settings.
28
References
Arulampalam, W., Gregg, P., and Gregory, M. 2001. “Unemployment Scarring.” The
Economic Journal 111: F577–F584.
Ask, H., Rognmo, K., Torvik, F., Røysamb, E., and Tambs, K. 2012. Non-Random Mating
and Convergence Over Time for Alcohol Consumption, Smoking, and Exercise: The Nord-
Trøndelag Health Study. Behavioral Genetics 42: 354-365.
Bargain, O., González, L., Keane, C., and Özcan, B. 2012. “Female Labor Supply and
Divorce: New Evidence from Ireland.” European Economic Review 56: 1675–1691.
Behrman, J., and Rosenzweig, M. 2004. “Return to Birthweight.” Review of Economics and
Statistics 86: 586–601.
Behrman, J. and Taubman, P. 1976. “Intergenerational Transmission of Income and
Wealth.” American Economic Review 66: 436–440.
Bemmels, H., Burt, A., Legrand, L., Iacono, W. and McGue, N. 2008. The Heritability of
Life Events: An Adolescent Twin and Adaption Study.” Twin Research and Human Genetics
11: 257–265.
Black, R., Devereux, P., and Salvanes, K. 2007. “From the Cradle to the Labor Market? The
Effect of Birth Weight on Adult Outcomes.” Quartely Journal of Economics 122: 409-439.
29
Black, R., Devereux, P., and Salvanes, K. 2015. “Losing Heart? The Effect of Job
Displacement on Health.” Industrial and Labor Relations Review 68: 833-861.
Boardman, J., Blalock, C. and Button, T. 2008. “Sex Differences in the Heritability of
Resilience.” Twin research and Human Genetics 11: 12–27.
Bound, J., and Solon, G. 1999. “Double Trouble: On the Value of Twins-based Estimation
of the Return to Schooling.” Economics of Education Review 18: 169–82.
Böckerman, P., Hyytinen, A. and Maczulskij, T. (2015a). Alcohol Consumption and Long-
Term Labor Market Outcomes.” Health Economics, Forthcoming.
Böckerman, P., Hyytinen, A. and Kaprio, J. (2015b). Smoking and Long-Term Labour
Market Outcomes.” Tobacco Control 24: 348–353.
Böckerman, P., and Ilmakunnas, P. 2009. “Unemployment and Self-Assessed Health:
Evidence from Panel Data”. Health Economics 18: 161-179.
Cherny, S., Cardon, L., Fulker, D. and DeFries, J. (1992). “Differential Heritability Across
Levels of Cognitive Ability.” Behavior Genetics 22: 153–162.
Clark, A., Diener, E., Georgellis, Y., and Lucas, R. 2008. “Lags and Leads in Life
Satisfaction: A Test of the Baseline Hypothesis.” Economic Journal 18: F222–F243.
30
Conger, R., Elder, G. Jr., Simons, R. and Ge, X. 1993. “Husband and Wife Differences in
Response to Undesirable Life Events.” Journal of Health and Social Behavior 34: 71–88.
Currie, J., and Madrian, B. 1999. “Health, Health Insurance and the Labor Market.” In
Handbook of Labor Economics, Ch. 50. Ashenfelter, O., and Card, D. (Eds.). Amsterdam:
North Holland, 3309-3407.
Dawson, D., Grant, B. and Ruan, J. 2005. “The Association between Stress and Drinking:
Modifying Effects of Gender and Vulnerability.” Alcohol and Alcoholism 40: 453–460.
DeFries, J. and Fulker, D. 1982. “Multiple Regression Analysis of Twin Data.” Behavior
Genetics 15: 467–473.
Doiron, D., and Mendolia, S. 2012. “The Impact of Job Loss on Family Dissolution.”
Journal of Population Economics 25: 367–398.
Eryigit-Madzwamuse, S., Strauss, V., Baumann, N., Bartmann, P. and Wolke, D. 2015.
“Personality of Adults who were born very preterm.” Archives of Disease in Childhood, doi:
10.1136/archdischild-2014-308007
Fletcher, J.M. 2013. “The Effects of Personality Traits on Adult Labor Market Outcomes:
Evidence from Siblings.” Journal of Economic Behavior and Organization 89: 122-35.
García-Gómez, P. 2011. “Institutions, Health Shocks and Labour Market Outcomes across
Europe.” Journal of Health Economics 30: 200–213.
31
Griliches, Z. 1979. “Siblings Models and Data in Economics: Beginnings of a Survey.”
Journal of Political Economy 87: S37–S64.
Haurin, D. 1989. “Women’s Labor Market Reactions to Family Disruptions.” Review of
Economics and Statistics 71: 54–61.
Holmes, T.H. and Rahe, R.H. (1967). “The Social Readjustment Rating Scale.” Journal of
Psychosomatic Research 11: 213–218.
Hyytinen, A., Ilmakunnas, P. and Toivanen, O. 2013. ”The Return-to-Entrepreneurship
Puzzle.” Labour Economics 20: 57–67.
Jacobson, L., LaLonde, R., and Sullivan, D. 1993. “Earning Losses of Displaced Workers.”
American Economic Review 83: 685–709.
Johnson, R., and Schoeni, R. 2011. “The Influence of Early-Life Events on Human Capital,
Health Status, and Labor Market Outcomes over the Life Course.” The B.E. Journal of
Economic Analysis & Policy 11. doi:10.2202/1935-1682.2521.
Johnson, W., and Skinner, J. 1986. “Labor Supply and Marital Separation.” The American
Economic Review 76: 455–469.
Kajantie, E., Hovi, P., Räikkönen, K., et al. 2008. “Young Adults with very Low Birth
Weight: Leaving the Parental Home and Sexual Relationships — Helsinki Study of Very
Low Birth Weight Adults.” Pediatrics 122: 62–72.
32
Kaprio, J. and Koskenvuo, M. 2002. “Genetic and Environmental Factors in Complex
Diseases: The Older Finnish Twin Cohort.” Twin Research 5: 358–365.
Kaprio, J., Koskenvuo, M., Artimo, M., Sarna, S. and Rantasalo, I. 1979. Baseline
characteristics of the Finnish Twin Registry: Section I: Materials, Methods,
Representativeness and Results for Variables Special to Twin Studies. Kansanterveystieteen
julkaisuja M47.
Kendler, K., Kuhn, J. and Prescott, C. 2012. ”The Interrelationship of Neuroticism, Sex, and
Stressful Life Events in the Prediction of Episodes of Major Depression.” American Journal
of Psychiatry 161: 631–636.
Kendler, K., Thornton, L. and Prescott, C. (1999). “Causal Relationship between Stressful
Life Events and the Onset of Major Depression.” American Journal of Psychiatry 156: 837–
841.
Kendler, K., Thornton, L. and Prescott, C. 2001. “Gender Differences in the Rates of
Exposure to Stressful Life Events and the Sensitivity to Their Depressogenic Effects.”
American Journal of Psychiatry 158: 587–593.
Kessler, R. and McLeod, J. 1984. “Sex Differences in Vulnerability to Undesirable Life
Events.” American Sociological Review 49: 620–631.
33
Kohler, H.P. and Rodgers, J. 2001. “DF-Like Analyses of Heritability with Double-Entry
Twin Data: Asymptotic Standard Errors and Efficient Estimation.” Behavior Genetics 31:
179–191.
Korkeamäki, O., and Kyyrä, T. 2014. “A Distributional Analysis of Earnings Losses of
Displaced Workers in an Economic Depression and Recovery.” Oxford Bulletin of
Economics and Statistics 76: 565–588.
Light, A., and Ahn, T. 2010. “Divorce as risky behavior.” Demography 47: 895–921.
Lillberg, K., Verkasalo, P.K., Kaprio, J., Teppo, L., Helenius, H., and Koskenvuo, M. 2003.
”Stressful Life Events and Risk of Breast Cancer in 10,808 Women: A Cohort Study.”
American Journal of Epidemiology, 157: 415–423.
Lucas, R., Diener, E., Gtob, A., Suh, E. and Shao, L. 2000. “Cross-Cultural Evidence for the
Fundamental Features of Extraversion.” Journal of Personality and Social Psychology 79:
452–468.
Lundborg, P., Nilsson, A., Rooth Dan-Olof. 2014. “Adolescent Health and Adult Labor
Market Outcomes.” Journal of Health Economics 37: 25–40.
Maczulskij, T. 2013a. “Employment Sector and Pay Gaps: Genetic and Environmental
Influences.” Labour Economics 23: 89–96.
34
Maczulskij, 2013b. “Economics of Wage Differentials and Public Sector Labour Markets.”
Jyväskylä Studies in Business and Economics No. 125.
.
Masuda, M. and Holmes, T. 1978. “Life-Events: Perceptions and Frequencies.”
Psychosomatic Medicine 40: 236–261.
Mckee, S.A., Maciejewski, P.K., Falba, T. and Mazure, CM. 2003. “Sex Differences in the
Effects of Stressful Life Events on Changes in Smoking Status.” Addiction 98: 847–855.
Misheva, V. 2015. “What Determines Emotional Well-Being? The Role of Adverse
Experiences: Evidence using Twin Data.” Journal of Happiness Studies, DOI:
10.1007/s10902-015-9678-9.
Mok, W., Meyer, B., Charles, K., and Achen, A. 2008. “A Note on “The Longitudinal
Structure of Earnings Losses among Work-Limited Disabled Workers”. Journal of Human
Resources 43: 721–728.
Nandi, A. and Nicoletti, C. 2014. “Explaining Personality Pay Gaps in the UK.” Applied
Economics 46: 3131–3150.
Riese, H., Snieder, H., Jeronimus, B., Korhonen, T., Rose, R., Kaprio, J. and Ormel, J. 2013.
“Timing of Stressful Life Events Affects Stability and Change of Neuroticism.” European
Journal of Personality 28: 193–200.
35
Schmitz, H. 2011. “Why are the Unemployed in Worse Health?” The causal effect of
unemployment on health. Labour Economics 18: 71–78.
“Measuring the Fffect of Husband’s Health on Wife's Labor Supply.” Health Economics 15:
579-601.
Silles, M. 2016. “The Impact of Children on Women’s Labour Supply and Earnings in the
UK: Evidence using Twin Births.” Oxford Economic Papers 68: 197–216.
Tennant, C. 2002. “Life Events, Stress and Depression: A Review of Recent Findings.”
Australian and New Zealand Journal of Psychiatry 36: 173–182.
Van den Berg, G., Lundborg, P., and Vikström, J. 2017. ”The Economics of Grief”. The
Economic Journal, 1-39.
Waaktaar, T. and Torgensen, S. 2012. “Genetic and Environmental Causes of Variation in
Trait Resilience in Young People.” Behavior Genetics 42: 366–377.
Waller, N. 1994. “A DeFries and Fulker Regression Model for Genetic Nonadditivity.”
Behavior Genetics 24: 149–153.
36
Tables and Figures
Figure 1. Histogram of the SLE index
37
Table 1. Prevalence of each stressful life events and SLE weights.
Prevalence SLEs, % SLE weight
Death of a spouse 2.7 0.973
Loss of a job 10.5 0.895
Divorce or separation 11.6 0.884
Marked increase in difficulties with spouse 15.1 0.849
Marked difficulties with boss or colleagues at work 19.2 0.808
Marked worsening in financial situation 20.2 0.798
Difficulties in sexual nature 21.9 0.781
Disease or injury causing over 3 weeks work disability 25.1 0.749
Marked change in the health of a family member 27.5 0.725
Death of a close relative or good friend 70.7 0.293
38
Table 2. Intra-class correlations and OLS estimates of DF-model.
Intra-class correlations OLS estimates of DF-model
DZ-twins MZ-twins Genetic Shared environment
SLE 0.13 *** 0.24 *** 0.234 (0.077) ***
0.005 (0.055)
SLE dependent financial 0.08 *** 0.34 *** 0.258 (0.033) ***
..
SLE dependent familial 0.10 *** 0.13 *** 0.120 (0.078)
0.038 (0.056)
SLE independent familial 0.08 *** 0.09 *** 0.009 (0.077) 0.112 (0.055) **
Notes: *** (p < 0.010), ** (p < 0.050)
39
Table 3. Summary statistics by gender
Men Women F-test
Within MZ
differences,
men
Within MZ
differences,
women
SLE
SLE index -0.051 0.038 12.37 *** 0.86 0.98
SLE, dep. work & financial 0.020 -0.015 1.85 0.79 0.75
SLE, dep. Familial -0.046 0.034 10.37 *** 0.87 0.93
SLE, indep. Familial -0.101 0.074 50.55 *** 0.86 0.92
Outcomes
Earnings, euros 23,969 17,145 553.57 *** 8,721 6,639
Social income transfers, euros 2,461 2,041 17.42 *** 2,622 2,221
Employment, months 10.01 9.53 37.65 *** 1.99 2.56
Basic controls
Age 43.2 42.1 35.13 *** 0 0
Education, years 12.0 11.8 8.00 *** 1.17 1.03
Married, dummy 0.80 0.80 0.19 0.19 0.24
No. of diseases in 1981 0.62 0.75 32.66 *** 0.71 0.72
Earnings in 1980, euros 20,230 11,991 886.48 *** 7,971 6,059
Mediators
Smoking, pack-years in 1990 10.13 4.00 446.79 *** 6.89 3.22
Passing out in 1990, dummy 0.20 0.07 219.65 *** 0.24 0.10
Tranquilizer use, dummy 0.09 0.13 33.51 *** 0.14 0.20
Extraversion 0.055 -0.041 14.56 *** 0.79 0.75
Neuroticism -0.101 0.074 48.29 *** 0.90 0.77
Number of obs. 2,732 3,700 483 700
Notes: Heteroscedasticity-robust F-test statistics for the null hypothesis of equal group
means. *** (p < 0.010). Within-MZ twin differences: the means of the absolute values of
the twin differences in the MZ sample.
40
Table 4. Regressions of long-term earnings, income transfers and employment for men.
All twins (1) DZ – MZ sample (2) DZ sample (3) MZ sample (4)
OLS regressions Twin-differences Twin-differences Twin-differences
Log(earnings)
A. SLE index -0.090 (0.017) *** -0.075 (0.022) *** -0.088 (0.029) *** -0.048 (0.028) *
B. SLE 3 classes:
SLE dep. work & financial -0.130 (0.016) *** -0.087 (0.020) *** -0.100 (0.026) *** -0.068 (0.026) ***
SLE dep. familial -0.004 (0.016) -0.022 (0.021) -0.041 (0.030) 0.016 (0.023)
SLE ind. familial 0.036 (0.015) ** 0.025 (0.018) 0.058 (0.024) ** -0.025 (0.025)
Log(income transfers)
A. SLE index 0.354 (0.048) *** 0.238 (0.070) *** 0.280 (0.082) *** 0.128 (0.130)
B. SLE 3 classes:
SLE dep. work & financial 0.379 (0.047) *** 0.240 (0.069) *** 0.236 (0.082) *** 0.206 (0.130)
SLE dep. familial 0.138 (0.054) ** 0.165 (0.076) ** 0.206 (0.096) ** 0.104 (0.129)
SLE ind. familial -0.119 (0.057) ** -0.182 (0.076) ** -0.165 (0.098) * -0.227 (0.125) *
Employment months
A. SLE index -0.474 (0.065) *** -0.331 (0.090) *** -0.337 (0.110) *** -0.307 (0.153) **
B. SLE 3 classes:
SLE dep. work & financial -0.636 (0.065) *** -0.476 (0.087) *** -0.489 (0.114) *** -0.460 (0.125) ***
SLE dep. familial -0.059 (0.066) -0.022 (0.090) 0.001 (0.116) -0.053 (0.139)
SLE ind. familial 0.171 (0.060) *** 0.139 (0.078) * 0.137 (0.100) 0.151 (0.120)
Number of obs. 2,732 1,366 883 483
Notes: Standard errors are robust to within-twin variation. *** (p < 0.010), ** (p < 0.050), * (p < 0.100). Additional controls include number of
chronic diseases, marital status, education years and previous earnings level. OLS specification in Colum1 also controls for age and age squared.
41
Table 5. Regressions of long-term earnings, income transfers and employment for women.
All twins (1) DZ – MZ sample (2) DZ sample (3) MZ sample (4)
OLS regressions Twin-differences Twin-differences Twin-differences
Log(earnings)
A. SLE index -0.037 (0.014) *** -0.040 (0.018) ** -0.064 (0.024) *** -0.001 (0.024)
B. SLE 3 classes:
SLE dep. work & financial -0.070 (0.014) *** -0.061 (0.018) *** -0.074 (0.023) *** -0.035 (0.025)
SLE dep. familial 0.027 (0.013) ** 0.014 (0.017) 0.009 (0.022) 0.039 (0.025)
SLE ind. familial -0.021 (0.013) -0.019 (0.018) -0.025 (0.024) -0.016 (0.026)
Log(income transfers)
A. SLE index 0.298 (0.041) *** 0.271 (0.064) *** 0.230 (0.086) *** 0.335 (0.091) ***
B. SLE 3 classes:
SLE dep. work & financial 0.178 (0.044) *** 0.186 (0.066) *** 0.183 (0.089) ** 0.188 (0.095) **
SLE dep. familial 0.122 (0.041) *** 0.054 (0.058) 0.049 (0.075) 0.050 (0.094)
SLE ind. familial 0.144 (0.041) *** 0.185 (0.059) *** 0.120 (0.076) 0.289 (0.095) ***
Empoyment months
A. SLE index -0.253 (0.059) *** -0.280 (0.078) *** -0.309 (0.104) *** -0.234 (0.116) **
B. SLE 3 classes:
SLE dep. work & financial -0.344 (0.062) *** -0.333 (0.077) *** -0.384 (0.099) *** -0.244 (0.120) **
SLE dep. familial -0.045 (0.059) -0.047 (0.076) -0.054 (0.099) -0.024 (0.116)
SLE ind. familial 0.043 (0.055) -0.022 (0.076) 0.014 (0.101) -0.094 (0.114)
Number of obs. 3,700 1,850 1,150 700
Notes: Standard errors are robust to within-twin variation. *** (p < 0.010), ** (p < 0.050), * (p < 0.100). Additional controls include number of
chronic diseases, marital status, education years and previous earnings level. OLS specification in Colum1 also controls for age and age squared.
42
Table 6. Within MZ twin-pair regressions of long-term earnings, social security benefits,
employment and self-employment: timing of SLE
(1) (2) (3)
Men Log(earnings)
Log(income
transfers) Employment
SLE, in the past 6 months -0.072 (0.026) *** 0.268 (0.123) ** -0.239 (0.126) *
SLE, in the past 5 years -0.035 (0.023) 0.145 (0.132) -0.222 (0.122) *
SLE, over 5 years ago 0.011 (0.025) -0.137 (0.123) -0.088 (0.152)
Other controls Yes Yes Yes
Women Log(earnings)
Log(income
transfers) Employment
SLE, in the past 6 months -0.004 (0.025) 0.242 (0.076) *** -0.093 (0.116)
SLE, in the past 5 years 0.013 (0.027) 0.294 (0.091) *** -0.219 (0.132) *
SLE, over 5 years ago -0.011 (0.024) 0.065 (0.090) -0.092 (0.102)
Other controls Yes Yes Yes
Notes: Standard errors are robust to within-twin variation. *** (p < 0.010), ** (p < 0.050), *
(p < 0.100). Number of observations: 483 for men and 700 for women. Other controls include
number of chronic diseases, marital status, education years, and previous earnings level.
43
Appendix
Table A1. Within-MZ correlations between stressful life events and individual
characteristics
Men SLE index
SLE, dep.
work &
financial
SLE,
indep.
Familial
SLE, dep.
Familial
No. of diseases, 1981 0.07 0.08 * -0.03 0.07
Smoking, pack-years in 1981 0.07 0.05 0.03 0.05
Alcohol use, 1981 0.11 ** 0.08 * 0.06 0.06
Extraversion, 1981 -0.003 0.002 0.06 -0.04
Neuroticism, 1981 0.23 *** 0.16 *** -0.01 0.22 ***
Wages in euros, 1980 0.05 0.01 -0.001 0.07
Unemployment, 1981 -0.004 0.02 -0.03 -0.001
Education years, 1981 0.01 -0.04 0.02 0.04
Women SLE index
SLE, dep.
work &
financial
SLE, dep.
Familial
SLE, indep.
Familial
No. of diseases, 1981 0.13 *** 0.12 *** 0.09 ** 0.08 **
Smoking, pack-years in 1981 0.14 *** 0.06 0.08 ** 0.14 ***
Alcohol use, 1981 0.08 ** -0.03 0.12 *** 0.08 **
Extraversion, 1981 -0.10 ** -0.10 *** -0.03 -0.07 *
Neuroticism, 1981 0.10 ** 0.07 * 0.05 0.08 **
Wages in euros, 1980 0.09 ** 0.01 0.02 0.13 ***
Unemployment, 1981 0.02 0.05 0.02 -0.01
Education years, 1981 -0.01 -0.03 -0.01 0.01